EN FR
Ship your build

Not sure you can do it? You already made a skill today.

Claude skills that work with Muse

Quick answer: Some Claude skills work with Meta Muse, and some do not. The ones that work share the open SKILL.md format: a short YAML header with a name and a description, followed by plain markdown instructions. When a skill is only instructions in that format, Muse can read it the same way another assistant can. When a skill depends on Claude-specific tools, commands, or bundled files that are not in its folder, it will not work as-is in Muse, even if the format looks familiar. Every Skill Harbor listing works with Muse today. The per-fiche "Works with" line is the source of truth, not this page.

What the open SKILL.md format is

A skill in this format is a folder with a SKILL.md file at its root. The file starts with a name and a description that tells the assistant when to use it, then gives instructions in markdown. Optional folders can sit beside it: scripts for repeatable work, references for documents loaded only when needed, and assets for files used in output. That is the whole format. It is readable by a human, reviewable before install, and not tied to one company's app by design. That is why a skill written for one assistant can sometimes run in another.

What transfers, and what does not

What usually transfers

  • Plain markdown instructions: steps, checklists, output templates, examples, and rules about what to ask before acting.
  • A clear trigger description: when the skill should run, and when it should not.
  • Standalone scripts that only use standard tools and files already listed in the skill folder.
  • Reference documents that are actually present in the skill folder.

What does NOT automatically transfer

  • Claude-specific wording and commands: instructions that say what "Claude" should do, Claude-only commands, or a test flow that assumes Claude's own testing tools.
  • Bundled files that are referenced but missing: if SKILL.md points to a reference, a script, or an agents file that is not in the folder, that part of the skill will fail in any assistant, including Muse.
  • Tool and MCP assumptions: a skill that assumes a specific MCP server, a specific CLI, or a specific model feature will only work where that tool exists.
  • Model-specific behaviour: a skill tuned for one model's habits may under-trigger or over-trigger in another model. That is a quality problem, not a format problem, and it has to be checked per skill.
  • Anything that needs an account, an API key, or a paid service: the skill itself may be free, but the service it connects to may not be. The listing's prerequisites line says so before you install.

The short version: the format transfers. The assumptions do not always transfer. That is why Skill Harbor checks per fiche instead of promising compatibility for a whole catalogue.

How to install a verified skill in Muse

Installs on Skill Harbor are copy-paste into Muse, never automatic. The steps are the same as for any listing:

  1. Open the listing and read the prerequisites line at the top of the install prompt. That line is the only thing you may have to do by hand, for example connecting an account or providing an API key in the secure way the prompt explains.
  2. Copy the install prompt in the Install section of the listing.
  3. Paste it into Muse and send it. Muse fetches the files, runs a safety check, installs the skill, and reports back what was installed and what remains for you, usually nothing.
  4. Start a new conversation to use the skill. Skills load when a conversation starts, so an existing chat will not see a newly installed skill.

Safety check, stated plainly: every install prompt includes a safety check. The agent scans for suspicious code and stops on a red flag instead of installing blindly. A verified creator badge confirms who the creator is, not that their code is safe, so the check still runs. Never paste passwords or live API keys into chat unless the prompt explains the secure way to provide them. The agent will never ask for your Skill Harbor login.

Verified examples, with links

These are the only examples this page claims. Each one was verified before drafting.

Replica Recon, Replica Build, and Replica Test: three steps of the Replica pack by Jakeschincariol, originally Claude skills, listed for Muse. Each is a free Development skill, pinned to commit 77c9436, and sandbox-tested on 2026-10-14 with no change detected on last check 2026-10-24. They show the transfer case at its cleanest: the same SKILL.md format, installed in Muse through the listing's copy-paste prompt.

Skill Creator by Anthropic: the Claude-origin skill-creator, already listed for Muse in the catalogue. It is included here as a catalogue example of a Claude-origin skill listed for Muse, not as a sandbox-tested claim. Its source is the Anthropic skills repository.

Quest Board: a Skill Harbor house build whose fiche already carries the site's own multi-agent badge, showing "Also works with: Grok, Claude, ChatGPT". That badge comes from the fiche's compatible_agents field, verified in D1 as grok,claude,chatgpt. It is included here to explain what that badge means, not as a promise that the Muse install prompt runs unchanged in Claude or ChatGPT.

What to check before trying a Claude skill elsewhere

If you find a Claude skill outside Skill Harbor and wonder whether it will work in Muse, check these five things before you paste anything:

  1. Is there a real SKILL.md file, with a name and a description, in a public repository with a README and a licence?
  2. Are all the files it references actually present in its folder?
  3. Does it mention Claude-specific commands, tools, or MCP servers that Muse does not have?
  4. Does it need an account, an API key, or a paid service, and is that stated up front?
  5. Does the Skill Harbor listing for it, if one exists, say it works with Muse? If there is no listing, there is no verified claim.

If any answer is no or unclear, do not assume it works. Ask on the listing, or use the Skill Harbor search to find a verified Muse listing for the same job.

Honesty guard, repeated on this page: the per-fiche "Works with" line is the source of truth, not this page. Not all skills transfer. This page explains the format and gives verified examples. It does not certify any skill that is not listed here as verified.

FAQ

Do Claude skills work with Muse?

Some do, some do not. Skills in the open SKILL.md format with no Claude-specific dependencies can work in Muse. The per-fiche "Works with" line is the source of truth, not this page.

How do I install a Claude-origin skill in Muse?

Open the Skill Harbor listing, copy the install prompt, paste it into Muse, and let Muse fetch, safety-check, install, and report back. Installs are copy-paste, never automatic.

Why does a Claude skill fail in Muse?

Usually because it references a missing bundled file, assumes a Claude-specific tool or MCP server, or uses a Claude-only command. Check the five points in the "What to check" section before retrying.

Does Skill Harbor guarantee that all Claude skills work in Muse?

No. Every Skill Harbor listing works with Muse today, but not every Claude skill in the world does. This page does not certify skills outside its verified examples.

What does "Also works with: Grok, Claude, ChatGPT" mean on a fiche?

It means the fiche's compatible_agents field lists those agents. It is the fiche's own claim, not a promise that the Muse install prompt runs unchanged in those agents.

Skill Harbor is an independent directory and is not affiliated with Meta.